Agents act as autonomous delegates, enabling scalable goal delegation and agent-to-agent competition that creates power asymmetries reshaping economic and socio-political outcomes.
Autonomous AI agents represent a significant technological evolution beyond current generative tools, acting as autonomous delegates that can perceive environments, plan, and execute complex, multi-step tasks independently. This shift from AI as a tool to AI as an autonomous actor introduces "agentic inequality"—the disparities in power, opportunity, and outcomes stemming from unequal access to and capabilities of AI agents. Unlike prior technologies that primarily augment human abilities, agents enable scalable delegation of goals and engage in direct agent-to-agent competition, creating novel power asymmetries that are poised to reshape economic and socio-political outcomes.
These dynamics manifest across three core dimensions: availability (who can access any agent), quality (differences in agent performance, intelligence, and tool access), and quantity (the ability to deploy swarms of coordinated agents). The combination of high-quality and numerous agents provides a synergistic advantage, enabling organizations or individuals to run millions of parallel simulations or automate complex workflows at scale—tasks infeasible for single or basic agents. This could accelerate the rise of "superstar firms" that leverage proprietary data to build superior agents, reinforcing competitive dominance through feedback loops of data and performance.
In consumer markets, individuals may be outmatched by sophisticated corporate agents in negotiations, as studies show advanced agents can consistently exploit less capable counterparts. In social and political domains, affluent individuals might deploy premium agents to navigate bureaucracies more effectively, while well-resourced actors could use agent swarms to dominate public consultations or generate personalized propaganda at unprecedented scale. These risks highlight how agentic inequality differs fundamentally from the traditional digital divide, evolving from mere access to tools into a structural imbalance of autonomous agency.
However, the technology also holds equalizing potential. If AI agents are treated as public infrastructure—universally accessible—they could democratize services, assist with legal or administrative processes, and empower consumers through automated comparison and fraud detection. The current period represents a window of opportunity to shape governance, investment, and standards to ensure equitable distribution of agentic power before inequalities become entrenched by default.
This research introduces the concept of "agentic inequality," a framework for understanding how the deployment of autonomous AI agents as delegates fundamentally alters power dynamics in digital and physical realms. Moving beyond simple automation, the paper explores agents as semi-independent actors capable of scalable goal delegation, where they can execute complex objectives and interact with other agents on behalf of human principals. The authors argue that this shift creates a new stratification layer in society, where access to superior agentic capabilities—measured by speed, strategic reasoning, and resource access—determines an entity's ability to compete. Unlike traditional digital tools, these agents can engage in agent-to-agent competition, effectively creating an economy where the efficiency of one's digital workforce dictates success.
The key contribution of this work is the analysis of how multi-agent interactions generate compounding power asymmetries. The study demonstrates that as agents compete for limited resources or informational advantages, small differences in underlying model capability or initial capital endowments can lead to massive, runaway disparities in outcomes. This is not merely a reflection of existing human inequality but an emergent property of autonomous systems optimizing for delegated goals. The paper details how these dynamics can reshape market structures, allowing those with sophisticated agentic networks to dominate economic activity and socio-political discourse through sheer scale and superior coordination.
This material is critical because it anticipates the destabilizing effects of an "agentic economy" that current economic and regulatory frameworks are ill-equipped to handle. It highlights the risk that unchecked agent competition could lead to systemic centralization of power, where a minority of actors control the most capable delegates, effectively disenfranchising those who cannot afford or access equivalent technology. By framing the issue around autonomy and delegation, the authors provide a necessary lens for policymakers and technologists to examine the long-term societal impacts of AI, urging the development of interventions that prevent agentic systems from exacerbating economic divides and undermining democratic processes.
The paper [Agentic Inequality](https://arxiv.org/html/2510.16853v2) (arXiv:2510.16853v2) explores the emergence of power asymmetries in systems where autonomous agents—whether software-based (e.g., LLMs, botnets) or human-like actors—compete for resources, influence, or goals. The authors argue that as agentic systems scale, they create agentic inequality: a dynamic where some agents accumulate disproportionate control over outcomes, analogous to economic or social inequality but with distinct mechanisms. Key contributions include:
1. Formalizing Agentic Power Dynamics: The paper introduces a framework to model how agents compete for resources (e.g., compute, attention, or data) and how these competitions lead to path-dependent amplification—where early advantages (e.g., superior algorithms, network effects) snowball into dominant positions. This challenges assumptions of "neutral" agent behavior, showing that autonomy + competition naturally produce winners and losers.
2. Socio-Political and Economic Implications: The authors warn that agentic inequality could exacerbate existing disparities, particularly in labor markets (e.g., AI-driven displacement) and governance (e.g., algorithmic influence over policy). They also highlight emergent hierarchies in multi-agent systems, where coalitions of agents (e.g., bot swarms) outcompete isolated actors, mimicking oligopolistic structures.
3. Mitigation Strategies: Proposals include procedural safeguards (e.g., resource quotas, decentralized training) and mechanism design (e.g., adversarial testing for power concentration) to curb runaway inequality. The paper stresses that these issues are not just technical but institutional, requiring policy interventions akin to antitrust or labor regulations.